Ian Duncan Examines Rationalist Networks and AI Doom Narratives

Summary

Ian Duncan’s essay examines the history of the rationalist and effective-altruist networks that helped shape contemporary AI-risk discourse and supplied personnel, funding, and ideas to AI laboratories, safety institutes, philanthropy, and government. It begins with Anthropic researcher Jacob Coxon’s September 2026 resignation, which warned that AI researchers sincerely believe advanced systems could cause human extinction; Anthropic alignment leader Evan Hubinger publicly agreed with the seriousness of that belief, while Elon Musk called the episode a psy-op. Duncan argues that assessing such warnings requires examining the subculture that produced many of its loudest advocates. He traces the movement from Eliezer Yudkowsky’s Singularity Institute, LessWrong, the Sequences, and Harry Potter and the Methods of Rationality to rituals such as Secular Solstice, and then to institutions including MIRI, CFAR, Oxford’s Future of Humanity Institute, and the Centre for Effective Altruism. The essay presents these networks as combining a shared apocalyptic worldview with dense overlaps in housing, romance, employment, philanthropy, and organizational leadership. It discusses allegations and admissions involving coercive psychological practices at CFAR and Leverage Research, the community’s response to abuse allegations, and the Zizian group, whose members have faced charges connected to multiple deaths and a 2025 Vermont shootout. Duncan also links AI-risk philanthropy to longtermist effective altruism, Open Philanthropy grants, Sam Bankman-Fried’s funding of Anthropic through FTX, Peter Thiel’s support for MIRI and Curtis Yarvin, and JD Vance’s stated intellectual connection to Yarvin. The essay emphasizes that these actors do not share a single political or regulatory position: Yudkowsky has advocated halting advanced AI development, while Thiel, Vance, and Marc Andreessen have opposed or criticized stronger regulation in different ways. It questions whether calls for licensing and external evaluation could entrench incumbent laboratories, citing concerns about METR’s disclosed relationships with frontier-lab personnel while acknowledging that the disclosure does not establish misconduct. Duncan’s conclusion is that AI warnings should neither be dismissed because of the networks behind them nor accepted solely because their advocates are sincere. He calls for enforceable conflict-of-interest rules, protected whistleblowers, independent replication, and oversight that also addresses discrimination, surveillance, labor, military use, and concentrated corporate power.